CCG-based Models for Statistical Machine Translation

The arguably best performing statistical machine translation systems are based on context-free formalisms or weakly equivalent ones. These models usually use a synchronous version of a context-free grammar (SCFG) which we argue is too rigid for the highly ambiguous task of human language translation. This is exacerbated by the fact that the imperfect methods available for aligning parallel texts make extracting an efficient grammar very hard. As a result, the context-free grammars extracted are usually very large in size after having already been restricted through a variety of constraints. We propose to use Combinatorial Categorial Grammar (CCG) for machine translation models. CCG is a lexicalized, mildly-context-sensitive formalism which is very well suited to capture long-distance dependencies that are not addressed very well by most current models. We believe that CCG is very well suited for the task of machine translation due to its ability to represent non-constituents in a syntactic way which frequently occur in parallel texts as well as its high derivational flexibility. This allows us to use some of the advantages of non-syntactic phrase-based approaches within a syntactic framework such as a relatively small grammar size compared to context-freebased machine translation grammars. A number of models leveraging the advantages of CCG are possible, however, our principal goal is to develop a string-to-tree based model which projects CCG on the target side of a synchronous grammar. We intend to apply the vast progress made in monolingual CCG parsing to machine translation. Additionally, we propose to extend CCG to a synchronous grammar (SCCG) as it has been done for other related formalisms such as tree adjoining grammars. We hope that a SCCG may provide similar derivational flexibility to monolingual CCG which may result in a better model for translational equivalence.